IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
Now that we actually know the cs timestamp frequency on gen4/5
let's run the corresponding test.
On g4x/ilk we must read the udw of the 64bit timestamp
register. Details in {g4x,gen5)_read_clock_frequency().
The one extra caveat is that on i965 (or at least CL, don't
recall if I ever tested on BW) we must read the register
twice to get an up to date value. For some unknown reason
the first read tends to return a stale value.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221031135703.14670-6-ville.syrjala@linux.intel.com
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
On ilk the UDW of TIMESTAMP increments every 1000 ns,
LDW is mbz. In order to represent that we'd need 52 bits,
but we only have 32 bits. Even worse most things want to
only deal with 32 bits of timestamp. So let's just set
up the timestamp frequency as if we only had the UDW.
On ctg/elk 63:20 of TIMESTAMP increments every 1/4 ns, 19:0
are mbz. To make life simpler let's ignore the LDW and set up
timestamp frequency based on the UDW only (increments every
1024 ns).
v2: Rebase
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221031135703.14670-2-ville.syrjala@linux.intel.com
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Update open coded for loop to use the standard scatterlist
for_each_sg API.
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221028155029.494736-4-matthew.auld@intel.com
Some minor cleanup of some variables for consistency.
Normalize struct sg_table to sgt.
Normalize struct dma_buf_attachment to attach.
checkpatch issues sizeof(), !NULL updates.
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221028155029.494736-3-matthew.auld@intel.com
Using PAGE_SIZE here potentially hides issues so bump that to something
larger. This should also make it possible for iommu to coalesce entries
for us. With that in place verify we can write from the GPU using the
importers sg_table, followed by checking that our writes match when read
from the CPU side.
v2: Switch over to igt_gpu_fill_dw(), which looks to be more widely
supported than the migrate stuff (at least OOTB).
References: https://gitlab.freedesktop.org/drm/intel/-/issues/7306
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Michael J. Ruhl <michael.j.ruhl@intel.com>
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221028155029.494736-2-matthew.auld@intel.com
We need to iterate over the original entries here for the sg_table,
pulling out the struct page for each one, to be remapped. However
currently this incorrectly iterates over the final dma mapped entries,
which is likely just one gigantic sg entry if the iommu is enabled,
leading to us only mapping the first struct page (and any physically
contiguous pages following it), even if there is potentially lots more
data to follow.
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/7306
Fixes: 1286ff739773 ("i915: add dmabuf/prime buffer sharing support.")
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Michael J. Ruhl <michael.j.ruhl@intel.com>
Cc: <stable@vger.kernel.org> # v3.5+
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221028155029.494736-1-matthew.auld@intel.com
We had already grabbed the rpm wakeref at obj destruction path,
but it also required to grab the wakeref when object moves.
When i915_gem_object_release_mmap_offset() gets called by
i915_ttm_move_notify(), it will release the mmap offset without
grabbing the wakeref. We want to avoid that therefore,
grab the wakeref at i915_ttm_unmap_virtual() accordingly.
While doing that also changed the lmem_userfault_lock from
mutex to spinlock, as spinlock widely used for list.
Also changed if (obj->userfault_count) to
GEM_BUG_ON(!obj->userfault_count).
v2:
- Removed lmem_userfault_{list,lock} from intel_gt. [Matt Auld]
Fixes: ad74457a6b5a ("drm/i915/dgfx: Release mmap on rpm suspend")
Suggested-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Anshuman Gupta <anshuman.gupta@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221027092242.1476080-3-anshuman.gupta@intel.com
Runtime pm is not really per GT, therefore it make sense to
move lmem_userfault_list, lmem_userfault_lock and
userfault_wakeref from intel_gt to intel_runtime_pm structure,
which is embedded to i915.
No functional change.
v2:
- Fixes the code comment nit. [Matt Auld]
Signed-off-by: Anshuman Gupta <anshuman.gupta@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Acked-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221027092242.1476080-2-anshuman.gupta@intel.com
The termination entries were missing for a couple of the recently-added
MTL steering tables.
Fixes: f32898c94a10 ("drm/i915/xelpg: Add multicast steering")
Fixes: a7ec65fc7e83 ("drm/i915/xelpmp: Add multicast steering for media GT")
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221028224022.964997-1-matthew.d.roper@intel.com
We have an additional register to select which slices contribute to
OAG/OAG counter increments.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221026222102.5526-16-umesh.nerlige.ramappa@intel.com
On DG2, a w/a resets RCS/CCS before it goes into RC6. This breaks OA
since OA does not expect engine resets during its use. Fix it by
disabling RC6.
v2: (Ashutosh)
- Bring back slpc_unset_param helper
- Update commit msg
- Use with_intel_runtime_pm helper for set/unset
v3: (Ashutosh)
- Just use intel_uc_uses_guc_rc
Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com>
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221026222102.5526-15-umesh.nerlige.ramappa@intel.com
If a drm client is killed, then hw contexts used by the client are reset
immediately. This reset clears the EU flex counter configuration. If an
OA use case is running in parallel, it would start seeing zeroed eu
counter values following the reset even if the drm client is restarted.
Save/restore the EU flex counter config so that the EU counters can be
monitored continuously across resets.
v2:
- Save/restore eu flex config only for gen12, as for pre-gen12, these
are saved and restored in the context image.
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221026222102.5526-14-umesh.nerlige.ramappa@intel.com
OA reports in the OA buffer contain an OA timestamp field that helps
user calculate delta between 2 OA reports. The calculation relies on the
CS timestamp frequency to convert the timestamp value to nanoseconds.
The CS timestamp frequency is a function of the CTC_SHIFT value in
RPM_CONFIG0.
In DG2, OA unit assumes that the CTC_SHIFT is 3, instead of using the
actual value from RPM_CONFIG0. At the user level, this results in an
error in calculating delta between 2 OA reports since the OA timestamp
is not shifted in the same manner as CS timestamp. Also the periodicity
of the reports is different from what the user configured because of
mismatch in the CS and OA frequencies.
The issue also affects MI_REPORT_PERF_COUNT command.
To resolve this, return actual OA timestamp frequency to the user in
i915_getparam_ioctl, so that user can calculate the right OA exponent as
well as interpret the reports correctly.
MR: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18893
v2:
- Use REG_FIELD_GET (Ashutosh)
- Update commit msg
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221026222102.5526-13-umesh.nerlige.ramappa@intel.com
Disable Clock gating in EU when gathering the events so that EU events
are not lost.
v2: Fix checkpatch issues
v3: User MCR helpers to write to MC reg
v4: Indent correctly (checkpatch)
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221026222102.5526-12-umesh.nerlige.ramappa@intel.com
DG2 introduces OA reports with 64 bit report header fields. Perf OA
would need more information about the OA format in order to process such
reports. Store all OA format info in oa_buffer instead of just the size
and format-id.
v2: Drop format_size variable (Ashutosh)
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221026222102.5526-11-umesh.nerlige.ramappa@intel.com
User passes uabi engine class and instance to the perf OA interface. Use
gt corresponding to the engine to pin the buffers to the right ggtt.
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221026222102.5526-10-umesh.nerlige.ramappa@intel.com
With multi-gt, user can access multiple OA buffers concurrently. Use
stream->lock instead of gt->perf.lock to serialize file operations.
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221026222102.5526-9-umesh.nerlige.ramappa@intel.com
Make perf part of gt as the OAG buffer is specific to a gt. The refactor
eventually simplifies programming the right OA buffer and the right HW
registers when supporting multiple gts.
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221026222102.5526-8-umesh.nerlige.ramappa@intel.com
Earlier code used exclusive_stream to check for user passed context.
Simplify this by accessing stream->ctx.
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221026222102.5526-7-umesh.nerlige.ramappa@intel.com
XEHPSDV and DG2 provide a way to configure bytes per clock vs commands
per clock reporting. Enable bytes per clock setting on enabling OA.
Bspec: 51762
Bspec: 52201
v2:
- Fix commit msg (Ashutosh)
- Fix checkpatch issues
v3:
- s/commands/bytes/ in code comment and commmit msg
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221026222102.5526-6-umesh.nerlige.ramappa@intel.com
Some SKUs of same gen12 platform may have different oactxctrl
offsets. For gen12, determine oactxctrl offsets at runtime.
v2: (Lionel)
- Move MI definitions to intel_gpu_commands.h
- Ensure __find_reg_in_lri does read past context image size
v3: (Ashutosh)
- Drop unnecessary use of double underscores
- fix find_reg_in_lri
- Return error if oa context offset is U32_MAX
- Error out if oa_ctx_ctrl_offset does not find offset
v4: (Ashutosh)
- Warn on odd MI LRI_LEN
- Remove unnecessary check for valid_oactxctrl_offset
- Drop valid_oactxctrl_offset macro
v5: Drop unrelated comment
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221026222102.5526-5-umesh.nerlige.ramappa@intel.com
Predication for batch buffer commands changed in XEHPSDV.
MI_BATCH_BUFFER_START predicates based on MI_SET_PREDICATE_RESULT
register. The MI_SET_PREDICATE_RESULT register can only be modified
with MI_SET_PREDICATE command. When configured, the MI_SET_PREDICATE
command sets MI_SET_PREDICATE_RESULT based on bit 0 of
MI_PREDICATE_RESULT_2. Use this to configure predication in noa_wait.
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221026222102.5526-4-umesh.nerlige.ramappa@intel.com
With GuC mode of submission, GuC is in control of defining the context
id field that is part of the OA reports. To filter reports, UMD and KMD
must know what sw context id was chosen by GuC. There is not interface
between KMD and GuC to determine this, so read the upper-dword of
EXECLIST_STATUS to filter/squash OA reports for the specific context.
v2: Explain guc id stealing w.r.t OA use case
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221026222102.5526-2-umesh.nerlige.ramappa@intel.com
When booting with CONFIG_CFI_CLANG, there are numerous violations when
accessing the files under
/sys/devices/pci0000:00/0000:00:02.0/drm/card0/gt/gt0:
$ cd /sys/devices/pci0000:00/0000:00:02.0/drm/card0/gt/gt0
$ grep . *
id:0
punit_req_freq_mhz:350
rc6_enable:1
rc6_residency_ms:214934
rps_act_freq_mhz:1300
rps_boost_freq_mhz:1300
rps_cur_freq_mhz:350
rps_max_freq_mhz:1300
rps_min_freq_mhz:350
rps_RP0_freq_mhz:1300
rps_RP1_freq_mhz:350
rps_RPn_freq_mhz:350
throttle_reason_pl1:0
throttle_reason_pl2:0
throttle_reason_pl4:0
throttle_reason_prochot:0
throttle_reason_ratl:0
throttle_reason_status:0
throttle_reason_thermal:0
throttle_reason_vr_tdc:0
throttle_reason_vr_thermalert:0
$ sudo dmesg &| grep "CFI failure at"
[ 214.595903] CFI failure at kobj_attr_show+0x19/0x30 (target: id_show+0x0/0x70 [i915]; expected type: 0xc527b809)
[ 214.596064] CFI failure at kobj_attr_show+0x19/0x30 (target: punit_req_freq_mhz_show+0x0/0x40 [i915]; expected type: 0xc527b809)
[ 214.596407] CFI failure at kobj_attr_show+0x19/0x30 (target: rc6_enable_show+0x0/0x40 [i915]; expected type: 0xc527b809)
[ 214.596528] CFI failure at kobj_attr_show+0x19/0x30 (target: rc6_residency_ms_show+0x0/0x270 [i915]; expected type: 0xc527b809)
[ 214.596682] CFI failure at kobj_attr_show+0x19/0x30 (target: act_freq_mhz_show+0x0/0xe0 [i915]; expected type: 0xc527b809)
[ 214.596792] CFI failure at kobj_attr_show+0x19/0x30 (target: boost_freq_mhz_show+0x0/0xe0 [i915]; expected type: 0xc527b809)
[ 214.596893] CFI failure at kobj_attr_show+0x19/0x30 (target: cur_freq_mhz_show+0x0/0xe0 [i915]; expected type: 0xc527b809)
[ 214.596996] CFI failure at kobj_attr_show+0x19/0x30 (target: max_freq_mhz_show+0x0/0xe0 [i915]; expected type: 0xc527b809)
[ 214.597099] CFI failure at kobj_attr_show+0x19/0x30 (target: min_freq_mhz_show+0x0/0xe0 [i915]; expected type: 0xc527b809)
[ 214.597198] CFI failure at kobj_attr_show+0x19/0x30 (target: RP0_freq_mhz_show+0x0/0xe0 [i915]; expected type: 0xc527b809)
[ 214.597301] CFI failure at kobj_attr_show+0x19/0x30 (target: RP1_freq_mhz_show+0x0/0xe0 [i915]; expected type: 0xc527b809)
[ 214.597405] CFI failure at kobj_attr_show+0x19/0x30 (target: RPn_freq_mhz_show+0x0/0xe0 [i915]; expected type: 0xc527b809)
[ 214.597538] CFI failure at kobj_attr_show+0x19/0x30 (target: throttle_reason_bool_show+0x0/0x50 [i915]; expected type: 0xc527b809)
[ 214.597701] CFI failure at kobj_attr_show+0x19/0x30 (target: throttle_reason_bool_show+0x0/0x50 [i915]; expected type: 0xc527b809)
[ 214.597836] CFI failure at kobj_attr_show+0x19/0x30 (target: throttle_reason_bool_show+0x0/0x50 [i915]; expected type: 0xc527b809)
[ 214.597952] CFI failure at kobj_attr_show+0x19/0x30 (target: throttle_reason_bool_show+0x0/0x50 [i915]; expected type: 0xc527b809)
[ 214.598071] CFI failure at kobj_attr_show+0x19/0x30 (target: throttle_reason_bool_show+0x0/0x50 [i915]; expected type: 0xc527b809)
[ 214.598177] CFI failure at kobj_attr_show+0x19/0x30 (target: throttle_reason_bool_show+0x0/0x50 [i915]; expected type: 0xc527b809)
[ 214.598307] CFI failure at kobj_attr_show+0x19/0x30 (target: throttle_reason_bool_show+0x0/0x50 [i915]; expected type: 0xc527b809)
[ 214.598439] CFI failure at kobj_attr_show+0x19/0x30 (target: throttle_reason_bool_show+0x0/0x50 [i915]; expected type: 0xc527b809)
[ 214.598542] CFI failure at kobj_attr_show+0x19/0x30 (target: throttle_reason_bool_show+0x0/0x50 [i915]; expected type: 0xc527b809)
With kCFI, indirect calls are validated against their expected type
versus actual type and failures occur when the two types do not match.
The ultimate issue is that these sysfs functions are expecting to be
called via dev_attr_show() but they may also be called via
kobj_attr_show(), as certain files are created under two different
kobjects that have two different sysfs_ops in intel_gt_sysfs_register(),
hence the warnings above. When accessing the gt_ files under
/sys/devices/pci0000:00/0000:00:02.0/drm/card0, which are using the same
sysfs functions, there are no violations, meaning the functions are
being called with the proper type.
To make everything work properly, adjust certain functions to match the
type of the ->show() and ->store() members in 'struct kobj_attribute'.
Add a macro to generate functions for that can be called via both
dev_attr_{show,store}() or kobj_attr_{show,store}() so that they can be
called through both kobject locations without violating kCFI and adjust
the attribute groups to account for this.
Link: https://github.com/ClangBuiltLinux/linux/issues/1716
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221013205909.1282545-1-nathan@kernel.org
We know that as long as GEM context create ioctl succeeds, a context was
created. There is no need to write about it, especially when such a message
heavily pollutes dmesg and makes debugging actual errors harder.
Since commit baa89ba3f1fe ("drm/i915/gem: initial conversion to new
logging macros using coccinelle"), the logging for creating a new user
context was moved under the driver debug output (for lack of a means for
per-user logs, and a lack of user-focused drm.debug parameter). This
only reveals how obnoxious having that spam be part of the driver debug
logs, so remove it. [ from Chris Wilson ]
Suggested-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Karolina Drobnik <karolina.drobnik@intel.com>
Cc: Andi Shyti <andi.shyti@linux.intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221025091903.986819-1-karolina.drobnik@intel.com
swiotlb_max_segment used to return either the maximum size that swiotlb
could bounce, or for Xen PV PAGE_SIZE even if swiotlb could bounce buffer
larger mappings. This made i915 on Xen PV work as it bypasses the
coherency aspect of the DMA API and can't cope with bounce buffering
and this avoided bounce buffering for the Xen/PV case.
So instead of adding this hack back, check for Xen/PV directly in i915
for the Xen case and otherwise use the proper DMA API helper to query
the maximum mapping size.
Replace swiotlb_max_segment() calls with dma_max_mapping_size().
In i915_gem_object_get_pages_internal() no longer consider max_segment
only if CONFIG_SWIOTLB is enabled. There can be other (iommu related)
causes of specific max segment sizes.
Fixes: a2daa27c0c61 ("swiotlb: simplify swiotlb_max_segment")
Reported-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com>
Signed-off-by: Robert Beckett <bob.beckett@collabora.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
[hch: added the Xen hack, rewrote the changelog]
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221020110308.1582518-1-hch@lst.de
The process for merging uAPI is to have UMD side ready and reviewed and
merged before merging. Revert for now until that is ready.
This reverts commit d54576a074a29d4901d0a693cd84e1a89057f694.
Reported-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: Michal Mrozek <michal.mrozek@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Stuart Summers <stuart.summers@intel.com>
Cc: Jordan Justen <jordan.l.justen@intel.com>
Cc: Yang A Shi <yang.a.shi@intel.com>
Cc: Nirmoy Das <nirmoy.das@intel.com>
Cc: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221024101946.28974-1-matthew.auld@intel.com
With the introduction of the delayed disable-sched behavior,
we use the GuC's xarray of valid guc-id's as a way to
identify if new requests had been added to a context
when the said context is being checked for closure.
Additionally that prior change also closes the race for when
a new incoming request fails to cancel the pending
delayed disable-sched worker.
With these two complementary checks, we see no more
use for intel_context:guc_state:number_committed_requests.
Signed-off-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221006225121.826257-3-alan.previn.teres.alexis@intel.com
Add a delay, configurable via debugfs (default 34ms), to disable
scheduling of a context after the pin count goes to zero. Disable
scheduling is a costly operation as it requires synchronizing with
the GuC. So the idea is that a delay allows the user to resubmit
something before doing this operation. This delay is only done if
the context isn't closed and less than a given threshold
(default is 3/4) of the guc_ids are in use.
Alan Previn: Matt Brost first introduced this patch back in Oct 2021.
However no real world workload with measured performance impact was
available to prove the intended results. Today, this series is being
republished in response to a real world workload that benefited greatly
from it along with measured performance improvement.
Workload description: 36 containers were created on a DG2 device where
each container was performing a combination of 720p 3d game rendering
and 30fps video encoding. The workload density was configured in a way
that guaranteed each container to ALWAYS be able to render and
encode no less than 30fps with a predefined maximum render + encode
latency time. That means the totality of all 36 containers and their
workloads were not saturating the engines to their max (in order to
maintain just enough headrooom to meet the min fps and max latencies
of incoming container submissions).
Problem statement: It was observed that the CPU core processing the i915
soft IRQ work was experiencing severe load. Using tracelogs and an
instrumentation patch to count specific i915 IRQ events, it was confirmed
that the majority of the CPU cycles were caused by the
gen11_other_irq_handler() -> guc_irq_handler() code path. The vast
majority of the cycles was determined to be processing a specific G2H
IRQ: i.e. INTEL_GUC_ACTION_SCHED_CONTEXT_MODE_DONE. These IRQs are sent
by GuC in response to i915 KMD sending H2G requests:
INTEL_GUC_ACTION_SCHED_CONTEXT_MODE_SET. Those H2G requests are sent
whenever a context goes idle so that we can unpin the context from GuC.
The high CPU utilization % symptom was limiting density scaling.
Root Cause Analysis: Because the incoming execution buffers were spread
across 36 different containers (each with multiple contexts) but the
system in totality was NOT saturated to the max, it was assumed that each
context was constantly idling between submissions. This was causing
a thrashing of unpinning contexts from GuC at one moment, followed quickly
by repinning them due to incoming workload the very next moment. These
event-pairs were being triggered across multiple contexts per container,
across all containers at the rate of > 30 times per sec per context.
Metrics: When running this workload without this patch, we measured an
average of ~69K INTEL_GUC_ACTION_SCHED_CONTEXT_MODE_DONE events every 10
seconds or ~10 million times over ~25+ mins. With this patch, the count
reduced to ~480 every 10 seconds or about ~28K over ~10 mins. The
improvement observed is ~99% for the average counts per 10 seconds.
Design awareness: Selftest impact.
As temporary WA disable this feature for the selftests. Selftests are
very timing sensitive and any change in timing can cause failure. A
follow up patch will fixup the selftests to understand this delay.
Design awareness: Race between guc_request_alloc and guc_context_close.
If a context close is issued while there is a request submission in
flight and a delayed schedule disable is pending, guc_context_close
and guc_request_alloc will race to cancel the delayed disable.
To close the race, make sure that guc_request_alloc waits for
guc_context_close to finish running before checking any state.
Design awareness: GT Reset event.
If a gt reset is triggered, as preparation steps, add an additional step
to ensure all contexts that have a pending delay-disable-schedule task
be flushed of it. Move them directly into the closed state after cancelling
the worker. This is okay because the existing flow flushes all
yet-to-arrive G2H's dropping them anyway.
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221006225121.826257-2-alan.previn.teres.alexis@intel.com
During GuC error capture initialization, we estimate the amount of size
we need for the error-capture-region of the shared GuC-log-buffer.
This calculation was incorrect so fix that. With the fixed calculation
we can reduce the allocation of error-capture region from 4MB to 1MB
(see note2 below for reasoning). Additionally, switch from drm_notice to
drm_debug for the 3X spare size check since that would be impossible to
hit without redesigning gpu_coredump framework to hold multiple captures.
NOTE1: Even for 1x the min size estimation case, actually running out
of space is a corner case because it can only occur if all engine
instances get reset all at once and i915 isn't able extract the capture
data fast enough within G2H handler worker.
NOTE2: With the corrected calculation, a DG2 part required ~77K and a PVC
required ~115K (1X min-est-size that is calculated as one-shot all-engine-
reset scenario).
Fixes: d7c15d76a554 ("drm/i915/guc: Check sizing of guc_capture output")
Cc: Alan Previn <alan.previn.teres.alexis@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Cc: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Cc: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Chris Wilson <chris.p.wilson@intel.com>
Signed-off-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221026060506.1007830-2-alan.previn.teres.alexis@intel.com
GuC will set the min/max frequencies to theoretical max on
ATS-M. This will break kernel ABI, so limit min/max frequency
to RP0(platform max) instead.
Also modify the SLPC selftest to update the min frequency
when we have a server part so that we can iterate between
platform min and max.
v2: Check softlimits instead of platform limits (Riana)
v3: More review comments (Ashutosh)
v4: No need to use saved_min_freq and other comments (Ashutosh)
Bug: https://gitlab.freedesktop.org/drm/intel/-/issues/7030
Acked-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Riana Tauro <riana.tauro@intel.com>
Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221024225453.4856-1-vinay.belgaumkar@intel.com
Waitboost (when SLPC is enabled) results in a H2G message. This can result
in thousands of messages during a stress test and fill up an already full
CTB. There is no need to request for boost if min softlimit is equal or
greater than it.
v2: Add the tracing back, and check requested freq
in the worker thread (Tvrtko)
v3: Check requested freq in dec_waiters as well
v4: Only check min_softlimit against boost_freq. Limit this
optimization for server parts for now.
v5: min_softlimit can be greater than boost (Ashutosh)
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221024171108.14373-1-vinay.belgaumkar@intel.com
If GuC is being used and we initialized GuC-error-capture,
we need to be warning if we don't provide an error-capture
register list in the firmware ADS, for valid GT engines.
A warning makes sense as this would impact debugability
without realizing why a reglist wasn't retrieved and reported
by GuC.
However, depending on the platform, we might have certain
engines that have a register list for engine instance error state
but not for engine class. Thus, add a check only to warn if the
register list was non existent vs an empty list (use the
empty lists to skip the warning).
NOTE: if a future platform were to introduce new registers
in place of what was an empty list on existing / legacy hardware
engines no warning is provided as the empty list is meant
to be used intentionally. As an example, if a future hardware
were to add blitter engine-class-registers (new) on top
of the legacy blitter engine-instance-register (HEAD, TAIL, etc.),
no warning is generated.
Signed-off-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221019072930.17755-2-alan.previn.teres.alexis@intel.com
A workaround was added to the driver to allow compute workloads to run
'forever' by disabling pre-emption on the RCS engine for Gen12.
It is not totally unbound as the heartbeat will kick in eventually
and cause a reset of the hung engine.
However, this does not work well in GuC submission mode. In GuC mode,
the pre-emption timeout is how GuC detects hung contexts and triggers
a per engine reset. Thus, disabling the timeout means also losing all
per engine reset ability. A full GT reset will still occur when the
heartbeat finally expires, but that is a much more destructive and
undesirable mechanism.
The purpose of the workaround is actually to give compute tasks longer
to reach a pre-emption point after a pre-emption request has been
issued. This is necessary because Gen12 does not support mid-thread
pre-emption and compute tasks can have long running threads.
So, rather than disabling the timeout completely, just set it to a
'long' value.
v2: Review feedback from Tvrtko - must hard code the 'long' value
instead of determining it algorithmically. So make it an extra CONFIG
definition. Also, remove the execlist centric comment from the
existing pre-emption timeout CONFIG option given that it applies to
more than just execlists.
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Acked-by: Michal Mrozek <michal.mrozek@intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221006213813.1563435-5-John.C.Harrison@Intel.com
Compute workloads are inherently not pre-emptible for long periods on
current hardware. As a workaround for this, the pre-emption timeout
for compute capable engines was disabled. This is undesirable with GuC
submission as it prevents per engine reset of hung contexts. Hence the
next patch will re-enable the timeout but bumped up by an order of
magnitude.
However, the heartbeat might not respect that. Depending upon current
activity, a pre-emption to the heartbeat pulse might not even be
attempted until the last heartbeat period. Which means that only one
period is granted for the pre-emption to occur. With the aforesaid
bump, the pre-emption timeout could be significantly larger than this
heartbeat period.
So adjust the heartbeat code to take the pre-emption timeout into
account. When it reaches the final (high priority) period, it now
ensures the delay before hitting reset is bigger than the pre-emption
timeout.
v2: Fix for selftests which adjust the heartbeat period manually.
v3: Add FIXME comment about selftests. Add extra FIXME comment and
drm_notices when setting heartbeat to a non-default value (review
feedback from Tvrtko)
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221006213813.1563435-4-John.C.Harrison@Intel.com
An earlier patch added support for compute engines. However, it missed
enabling the anti-pre-emption w/a for the new engine class. So move
the 'compute capable' flag earlier and use it for the pre-emption w/a
test.
Fixes: c674c5b9342e ("drm/i915/xehp: CCS should use RCS setup functions")
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Aravind Iddamsetty <aravind.iddamsetty@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Cc: Jason Ekstrand <jason@jlekstrand.net>
Cc: "Michał Winiarski" <michal.winiarski@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tejas Upadhyay <tejaskumarx.surendrakumar.upadhyay@intel.com>
Cc: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Cc: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>
Cc: Stuart Summers <stuart.summers@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Ramalingam C <ramalingam.c@intel.com>
Cc: Akeem G Abodunrin <akeem.g.abodunrin@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221006213813.1563435-3-John.C.Harrison@Intel.com
GuC converts the pre-emption timeout and timeslice quantum values into
clock ticks internally. That significantly reduces the point of 32bit
overflow. On current platforms, worst case scenario is approximately
110 seconds. Rather than allowing the user to set higher values and
then get confused by early timeouts, add limits when setting these
values.
v2: Add helper functions for clamping (review feedback from Tvrtko).
v3: Add a bunch of BUG_ON range checks in addition to the checks
already in the clamping functions (Tvrtko)
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221006213813.1563435-2-John.C.Harrison@Intel.com
This patch replaces all occurences of the form
intel_uncore_write(reg, intel_uncore_read(reg) OP val)
with intel_uncore_rmw.
Signed-off-by: Andrzej Hajda <andrzej.hajda@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221019143818.244339-2-andrzej.hajda@intel.com
This patch replaces all occurences of the form
intel_uncore_write(reg, intel_uncore_read(reg) OP val)
with intel_uncore_rmw.
Signed-off-by: Andrzej Hajda <andrzej.hajda@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221019143818.244339-1-andrzej.hajda@intel.com
A misplaced closing parenthesis caused the groupid/instanceid values to
be considered part of the ternary operator's condition instead of being
OR'd into the resulting value.
Fixes: f32898c94a10 ("drm/i915/xelpg: Add multicast steering")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Arun R Murthy <arun.r.murthy@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221019222437.3035182-1-matthew.d.roper@intel.com
Since a7c01fa93aeb ("signal: break out of wait loops on kthread_stop()")
kthread_stop() started asserting a pending signal which wreaks havoc with
a few of our selftests. Mainly because they are not fully expecting to
handle signals, but also cutting the intended test runtimes short due
signal_pending() now returning true (via __igt_timeout), which therefore
breaks both the patterns of:
kthread_run()
..sleep for igt_timeout_ms to allow test to exercise stuff..
kthread_stop()
And check for errors recorded in the thread.
And also:
Main thread | Test thread
---------------+------------------------------
kthread_run() |
kthread_stop() | do stuff until __igt_timeout
| -- exits early due signal --
Where this kthread_stop() was assume would have a "join" semantics, which
it would have had if not the new signal assertion issue.
To recap, threads are now likely to catch a previously impossible
ERESTARTSYS or EINTR, marking the test as failed, or have a pointlessly
short run time.
To work around this start using kthread_work(er) API which provides
an explicit way of waiting for threads to exit. And for cases where
parent controls the test duration we add explicit signaling which threads
will now use instead of relying on kthread_should_stop().
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221020130841.3845791-1-tvrtko.ursulin@linux.intel.com
